Abstract: The reliability of deep excavation numerical simulation results hinges on the authenticity of the geological model. Current homogeneous 3D models often lead to prediction errors because they neglect spatial variability in soil properties. This paper proposes a refined method integrating CAD intelligent parsing with 3D geological modeling. An automated borehole data extraction algorithm is developed to intelligently identify complex geological features such as stratigraphic discontinuities and lenses. Furthermore, a semi-automated workflow bridging Civil 3D and PLAXIS 3D is established, achieving high-fidelity conversion from structured data to 3D finite element models. Analysis of the Suzhou River Deep Tunnel Project excavation pit demonstrates that, compared to traditional uniform-thickness models, the refined variable-thickness 3D model more accurately reveals the excavation pit-soil interaction mechanism. The error in predicting the surface deformation influence zone outside the pit reaches 11.75%, while the maximum error in calculating internal forces within the support system reaches 15.12%. This study shows that the method enables high-precision excavation pit simulation driven by 3D geological modeling.